Datasets:

ArXiv:
License:
adwaitagashe commited on
Commit
7e878ad
·
1 Parent(s): f06e4c9

added viewpoint filtering

Browse files
Files changed (1) hide show
  1. bordirlines.py +11 -10
bordirlines.py CHANGED
@@ -95,11 +95,13 @@ class BordIRLinesDataset(datasets.GeneratorBasedBuilder):
95
  for lang in SUPPORTED_LANGUAGES
96
  ]
97
 
98
- def __init__(self, *args, relevance_filter="all", annotation_type=None, llm_mode="fewshot", **kwargs):
99
  super().__init__(*args, **kwargs)
100
  self.relevance_filter = relevance_filter # "relevant", "non-relevant", or "all"
101
  self.annotation_type = annotation_type
102
  self.llm_mode = llm_mode # Default to "fewshot"
 
 
103
 
104
  def _info(self):
105
  return datasets.DatasetInfo(
@@ -116,7 +118,7 @@ class BordIRLinesDataset(datasets.GeneratorBasedBuilder):
116
  "doc_text": datasets.Value("string"),
117
  "doc_lang": datasets.Value("string"),
118
  "relevant_human": datasets.Value("bool"),
119
- "territory_human": datasets.Value("string"),
120
  "relevant_llm_zeroshot": datasets.Value("bool"),
121
  "relevant_llm_fewshot": datasets.Value("bool"),
122
  }
@@ -196,11 +198,8 @@ class BordIRLinesDataset(datasets.GeneratorBasedBuilder):
196
  # Get Human Data
197
  human_data = human_map.get((query_id, doc_id), {})
198
 
199
- # Directly use the new 'relevant' column (no need for tie-breaking)
200
- relevant_human = human_data.get("relevant", False) # Default to False if missing
201
-
202
- # Directly use the 'territory' column instead of processing 'territory_human'
203
- territory_human = human_data.get("territory", "")
204
 
205
  # Get LLM Data
206
  llm_data = llm_map.get((query_id, doc_id), {})
@@ -209,7 +208,9 @@ class BordIRLinesDataset(datasets.GeneratorBasedBuilder):
209
  if self.llm_mode == "fewshot"
210
  else llm_data.get("relevant_zeroshot", None)
211
  )
212
-
 
 
213
  # Filtering logic based on relevance preference
214
  if self.relevance_filter == "relevant":
215
  if self.annotation_type == "human" and not relevant_human:
@@ -242,9 +243,9 @@ class BordIRLinesDataset(datasets.GeneratorBasedBuilder):
242
  "doc_text": docs[doc_lang][doc_id],
243
  "doc_lang": doc_lang,
244
  "relevant_human": relevant_human,
245
- "territory_human": territory_human,
246
  "relevant_llm_zeroshot": llm_data.get("relevant_zeroshot", None),
247
  "relevant_llm_fewshot": llm_data.get("relevant_fewshot", None),
248
  },
249
  )
250
- counter+=1
 
95
  for lang in SUPPORTED_LANGUAGES
96
  ]
97
 
98
+ def __init__(self, *args, relevance_filter="all", annotation_type=None, llm_mode="fewshot", viewpoint_filter=None, **kwargs):
99
  super().__init__(*args, **kwargs)
100
  self.relevance_filter = relevance_filter # "relevant", "non-relevant", or "all"
101
  self.annotation_type = annotation_type
102
  self.llm_mode = llm_mode # Default to "fewshot"
103
+ self.viewpoint_filter = viewpoint_filter # Filter for a specific viewpoint
104
+
105
 
106
  def _info(self):
107
  return datasets.DatasetInfo(
 
118
  "doc_text": datasets.Value("string"),
119
  "doc_lang": datasets.Value("string"),
120
  "relevant_human": datasets.Value("bool"),
121
+ "viewpoint": datasets.Value("string"),
122
  "relevant_llm_zeroshot": datasets.Value("bool"),
123
  "relevant_llm_fewshot": datasets.Value("bool"),
124
  }
 
198
  # Get Human Data
199
  human_data = human_map.get((query_id, doc_id), {})
200
 
201
+ relevant_human = human_data.get("relevant", False)
202
+ viewpoint_human = human_data.get("territory", "")
 
 
 
203
 
204
  # Get LLM Data
205
  llm_data = llm_map.get((query_id, doc_id), {})
 
208
  if self.llm_mode == "fewshot"
209
  else llm_data.get("relevant_zeroshot", None)
210
  )
211
+ viewpoint = viewpoint_human
212
+ if self.viewpoint_filter and self.viewpoint_filter not in viewpoint:
213
+ continue
214
  # Filtering logic based on relevance preference
215
  if self.relevance_filter == "relevant":
216
  if self.annotation_type == "human" and not relevant_human:
 
243
  "doc_text": docs[doc_lang][doc_id],
244
  "doc_lang": doc_lang,
245
  "relevant_human": relevant_human,
246
+ "viewpoint": viewpoint,
247
  "relevant_llm_zeroshot": llm_data.get("relevant_zeroshot", None),
248
  "relevant_llm_fewshot": llm_data.get("relevant_fewshot", None),
249
  },
250
  )
251
+ counter += 1